The Best Open Source FinOps Tools in 2026

The open source FinOps tools worth running in 2026 are OpenCost for Kubernetes cost allocation, Cloud Custodian for cost policies as code, Infracost for the cost of an infrastructure change before it merges, OptScale for a self-hosted FinOps platform, OpenOps for approved cost workflows, and Steampipe with Turbot's Thrifty mods for SQL waste checks. Add Microsoft's FinOps toolkit on Azure or AWS's Cloud Intelligence Dashboards on AWS for reporting, and Karpenter, KRR, and Goldilocks to right-size Kubernetes. Each does one job well, costs nothing to license, and asks you to run it.

We read each project's repository, docs, and release history on October 26, 2026, and kept only projects with a release or steady commits in 2026. For every tool we note what it does, whether it acts or only reports, its license, and what it takes to run. We make LevelFour, a commercial platform, so it stays off the list and appears once, near the end.

What makes an open source FinOps tool worth running?

Is it still maintained?

Stars measure attention, not upkeep. Check the date of the last release and the commits of the last 90 days. A cost tool reads billing formats and cloud APIs that change every year, and a project that stopped shipping falls behind them.

What does the license let you do?

Apache-2.0 and MIT let you run and change the code with few conditions. AGPL-3.0 asks you to share changes when you offer the software as a service. A source-available license, such as the Elastic License 2.0, is not open source, even when the code sits on GitHub.

Does it act, or only report?

Most tools show where the money goes. Fewer change anything: Cloud Custodian runs policy actions, Karpenter replaces nodes, and OpenOps runs workflows after an approval. Decide which you need before you compare features.

What does it cost to run?

The license is free; the infrastructure and the engineer who keeps it running are not. Microsoft estimates FinOps hubs from $120 a month, and AWS's Cloud Intelligence Dashboards project estimates its foundational dashboards at $100 to $200 a month for a sample setup. Count the hours too.

Does it read FOCUS?

FOCUS is the FinOps Foundation's open specification for billing data, at version 1.4 since June 2026. A tool that reads FOCUS can combine bills from several clouds in one schema. FOCUS specification.

The eleven tools at a glance

1. OpenCost

OpenCost measures and allocates Kubernetes and cloud infrastructure costs. Kubecost built and open sourced it, and it has been a CNCF Incubating project since October 2024. OpenCost at the CNCF.

• Cost allocation in real time by cluster, node, namespace, controller, service, or pod, covering CPU, GPU, memory, and persistent volumes. OpenCost on GitHub.
• Cloud cost monitoring for AWS, Azure, Google Cloud, and Oracle Cloud, with on-premises clusters priced from a CSV file. Configuration.
• Prometheus metrics, a cost API, kubectl cost, a web UI, and an MCP server you can switch on.
• Plugins that bring Datadog, MongoDB Atlas, and OpenAI costs in, with responses in the FOCUS format. OpenCost plugins.

License and upkeep: Apache-2.0. Version 1.121.3 shipped on September 16, 2026, and the repository took 56 commits in the 90 days to October 26. You install it with the official Helm chart.

IBM Kubecost is the commercial product from the team that built OpenCost. Its Foundations tier is free for unlimited clusters up to 250 cores, and IBM's Kubecost 3.0 announcement sets the free tier by spend instead, up to $100,000 over 30 days. IBM Kubecost · Kubecost 3.0.

Best for: platform teams that want Kubernetes cost by namespace, workload, and label as a free, vendor-neutral standard.

2. Cloud Custodian

Cloud Custodian is a rules engine for cloud accounts: you write YAML policies that find resources and act on them. It has been a CNCF Incubating project since September 2022. Cloud Custodian at the CNCF.

• Policies built from filters and actions on a resource type, with a dry-run mode that shows what a policy would do. Cloud Custodian on GitHub.
• Cost actions such as off-hours schedules for EC2 instances and Auto Scaling groups, and a notify, stop, then delete workflow for unused RDS databases. Off-hours · Unused RDS.
• Runs on a schedule, or in real time from cloud events it sets up itself, across many accounts, subscriptions, and projects.
• Policies run against infrastructure as code in CI, and c7n_mailer sends the notifications.

License and upkeep: Apache-2.0. Release 0.9.53.0 shipped on October 1, 2026, with more than 100 commits in 90 days. You install it with pip or the Docker image. It manages AWS, Azure, and Google Cloud, with Kubernetes, Tencent Cloud, and OpenStack in beta.

Best for: teams that already treat governance as code and want off-hours schedules, idle cleanup, and tag rules enforced the same way.

3. Infracost

Infracost estimates what a change to infrastructure code will cost before you deploy it, and checks it against FinOps policies. It reads Terraform, Terragrunt, CloudFormation, and AWS CDK. Infracost on GitHub.

• A cost breakdown per resource for a project, and a cost diff with policy checks on each pull request, from CI on GitHub, GitLab, or Azure DevOps. CI/CD integrations.
• More than 1,100 resource types across AWS, Azure, and Google Cloud, plus estimates for usage-based resources such as S3 and Lambda.
• Editor extensions for VS Code, Cursor, JetBrains IDEs, Neovim, and Zed, built on its language server, and agent skills that let Claude Code and Cursor check cost and tagging policies while they write infrastructure code. Agent skills.
• Tagging policies and guardrails, defined once in Infracost Cloud and enforced in the CLI, the editors, the agents, and CI. Guardrails.

License and upkeep: Apache-2.0. Infracost 2.0 split the code into focused repositories; the CLI, in infracost/cli, shipped version 2.18.3 on October 28, 2026, with 25 commits in 90 days. You install it with Homebrew, a script, or Chocolatey, and authenticate with infracost setup.

Infracost Cloud is the paid service: CI/CD is free up to 1,000 runs a month, Starter costs $250 a month for 10,000 runs, Cloud costs $1,000 a month with 10 admin seats, and Enterprise is by quote. Infracost pricing.

Best for: teams that ship infrastructure as code and want each change's cost in the pull request before it merges.

4. OptScale

OptScale, from Hystax, is a FinOps platform you host yourself, for AWS, Azure, Google Cloud, Alibaba Cloud, and Kubernetes. OptScale on GitHub.

• Idle and unused resource detection, and rightsizing recommendations.
• Power schedules that stop non-production environments outside working hours.
• Analysis of Reserved Instances, Savings Plans, and Spot.
• Dashboards, budgets, anomaly alerts, and ownership views, plus Databricks cost analytics and Kubernetes allocation by namespace, workload, and label.

License and upkeep: Apache-2.0. A release shipped on October 12, 2026, with 74 commits in 90 days. A Docker Compose mode lets you try it; production runs on Kubernetes and asks for 8 or more cores, 16 GB of RAM, and 150 GB of SSD.

Hystax also sells it hosted: free for cloud accounts under $5,000 of monthly spend, Pro at $95 per cloud account a month, and Enterprise by quote. OptScale pricing.

Best for: teams that want a full FinOps platform across several clouds and prefer to host it themselves.

5. OpenOps

OpenOps is a no-code platform for FinOps workflows: it takes savings opportunities and runs each one through an approval to an action. OpenOps on GitHub.

• A library of ready workflows for cost optimization, tagging, budgeting, allocation, and reporting.
• A no-code editor that lets you drop into code when a step needs it.
• Human approvals across channels, with versioned workflows and logs.
• Tables where you approve, dismiss, or snooze opportunities gathered from native and third-party FinOps tools.

License and upkeep: Apache-2.0. Version 0.7.1 shipped on September 24, 2026, with more than 100 commits in 90 days. It installs free with Docker Compose and works with AWS, Azure, and Google Cloud. A managed service with support is sold separately. OpenOps pricing.

Best for: FinOps practitioners who want recommendations turned into approved, logged actions without writing scripts.

6. Steampipe and Powerpipe with the Thrifty mods

Steampipe exposes cloud APIs as SQL tables, Powerpipe runs dashboards and benchmarks on them, and Turbot's Thrifty mods are ready-made waste checks. AWS Thrifty mod.

• AWS checks on EBS, EC2, RDS, S3, Lambda, CloudWatch, EKS, and a dozen more, including checks for unused EBS and RDS snapshots.
• Thrifty mods for Azure, Google Cloud, Oracle Cloud, DigitalOcean, and Alibaba Cloud.
• Results as a local dashboard or as checks in a terminal, and cost tables such as daily cost by service and Cost Optimization Hub recommendations in the AWS plugin. Steampipe AWS plugin.
• Flowpipe Thrifty mods that detect and correct, resource by resource with an approval, as a notification only, or on their own. Flowpipe AWS Thrifty.

License and upkeep: Steampipe and Powerpipe are AGPL-3.0, and the mods and the AWS plugin are Apache-2.0. Steampipe 2.4.8 and Powerpipe 1.5.6 shipped on October 22, 2026, and AWS Thrifty 1.2.0 on May 28, 2026. You install them with Homebrew. Turbot Pipes is the hosted version.

Best for: engineers who know SQL and want quick waste checks across clouds, defined as code.

7. Microsoft FinOps toolkit

Microsoft's FinOps toolkit is an open source set of tools for FinOps on the Microsoft Cloud. Its center is FinOps hubs, a cost data platform aligned with FOCUS. FinOps toolkit.

• Reporting across accounts and subscriptions in separate tenants, with negotiated and commitment discount savings for EA and MCA. FinOps hubs.
• Data in Azure Data Explorer or Microsoft Fabric, five Power BI report templates, and an MCP server for AI tools such as GitHub Copilot.
• FOCUS data from AWS, Google Cloud, Alibaba Cloud, Oracle Cloud, and Tencent Cloud, once you copy it into the hub. Hub data model.
• Workbooks, the Azure Optimization Engine, a PowerShell module, and open data on pricing units, regions, and services.

License and upkeep: MIT. Version 14 shipped in April 2026, and the repository took 63 commits in 90 days. You deploy it from an Azure template. Microsoft estimates a hub from $120 a month plus $10 a month per $1 million of cost it monitors. Hub costs.

Best for: Azure-first organizations on EA or MCA that want a FOCUS cost platform and Power BI reports they own.

8. AWS Cloud Intelligence Dashboards

The Cloud Intelligence Dashboards, CUDOS among them, are an open source framework maintained by AWS staff: more than 20 QuickSight dashboards on your own cost and usage data. Cloud Intelligence Dashboards on GitHub.

• Foundational dashboards, among them CUDOS, the Cost Intelligence Dashboard, and KPI and Modernization, that need only the Cost and Usage Report 2.0.
• Advanced dashboards that add Trusted Advisor, Compute Optimizer, Cost Anomaly Detection, and Budgets data.
• A FOCUS dashboard, a dashboard of Cost Optimization Hub recommendations, and dashboards for Azure and Google Cloud data. The dashboards.

License and upkeep: MIT-0. Version 4.4.18 shipped on October 9, 2026, with 36 commits in 90 days. You deploy it with the cid-cmd CLI or CloudFormation. The project's README estimates the foundational dashboards at $100 to $200 a month in S3, Glue, Athena, and QuickSight for a sample setup.

Best for: AWS organizations that want ready executive and engineering dashboards on their own CUR data.

9. Karpenter

Karpenter is a Kubernetes node autoscaler: it launches nodes that fit pending pods and removes nodes the cluster no longer needs. It is a Kubernetes SIG Autoscaling project. Karpenter on GitHub.

• Consolidation that removes nodes whose pods fit elsewhere and replaces nodes with cheaper ones. Disruption.
• Spot support, with deletion consolidation on by default and Spot-to-Spot replacement behind a feature flag.
• Providers for AWS and Azure, one for Oracle Cloud supported by Oracle, and community providers for Google Cloud and others.

License and upkeep: Apache-2.0. Version 1.15.1 shipped on October 23, 2026, with more than 100 commits in 90 days. You install it in the cluster with Helm.

Best for: Kubernetes teams that want node savings from bin-packing, consolidation, and Spot without tuning node groups by hand.

10. KRR

KRR, from Robusta, is a command-line tool that reads pod usage from Prometheus and recommends CPU and memory requests and limits. KRR on GitHub.

• No agent in the cluster: it runs from a laptop against your current kube context, or in the cluster for weekly Slack reports.
• Reads Prometheus, Thanos, VictoriaMetrics, Grafana Mimir, and the managed Prometheus services of Amazon, Azure, and Google.
• A default strategy that sets the CPU request at the 95th percentile with no limit, and memory at its observed peak plus 15%.
• Output as a table, JSON, YAML, CSV, or HTML, and an Enforcer that applies the recommendations once Robusta, which bundles KRR, runs in the cluster with a Robusta account (free tier available). KRR Enforcer.

License and upkeep: MIT. Version 1.31.1 shipped on October 19, 2026. You install it with Homebrew, Docker, or a binary. Robusta's SaaS adds a UI.

Best for: teams that already run Prometheus and want rightsizing numbers today, without installing anything in the cluster.

11. Goldilocks

Goldilocks, from Fairwinds, creates a Vertical Pod Autoscaler in recommendation mode for each workload and shows the suggested requests and limits on a dashboard. Goldilocks on GitHub.

• Opt-in per namespace with one label. Installation.
• Needs only the VPA recommender and metrics-server.
• A dashboard per namespace with the suggested requests and limits for each workload.

License and upkeep: Apache-2.0. Version 4.16.2 shipped on September 15, 2026. You install it with a Helm chart. Fairwinds Insights adds multi-cluster views and history.

Best for: teams that want a visual, per-namespace starting point for requests and limits.

Which open source tools track AI and LLM costs?

The tools above read AI spend that bills through a cloud, such as Bedrock on the AWS bill. For model APIs you call directly, these projects count the tokens and the money:

• LiteLLM, an open source gateway for more than 100 models, sets budgets per key, user, team, and customer, and stops a key's requests once it passes its budget. MIT outside its enterprise folder. LiteLLM budgets.
• Langfuse tracks the usage and cost of every LLM call, with dashboards and spend alerts. MIT outside its enterprise folders; you host it or use Langfuse Cloud. Langfuse cost tracking.
• OpenCost's OpenAI plugin, and its inference cost tracking for vLLM deployments, which reports cost per million tokens.

How to choose

1. Start from the job

Kubernetes allocation points to OpenCost. Cost rules as code point to Cloud Custodian, and the cost of a change before it merges to Infracost. A whole platform you host points to OptScale. Reports on one cloud point to Microsoft's toolkit or AWS's dashboards. Kubernetes savings point to Karpenter, KRR, and Goldilocks, and approved actions to OpenOps.

2. Check the upkeep and the license

Look for a release in the last few months and commits in the last 90 days. Read the license before you change the code or offer it to others as a service.

3. Count the running cost

Add the cloud resources the tool runs on to the hours of the engineer who upgrades it. A hub on Azure, dashboards on AWS, or an OptScale cluster each carry a monthly bill.

4. Decide who acts

Reports need someone to make each change. If nobody owns that work, pick a tool that acts, and keep its changes in your infrastructure code so the next deploy doesn't undo them.

5. Check the bill

An estimate is a forecast. Compare the next month's bill with the last one, resource by resource, before you count a saving.

Common mistakes with open source FinOps tools

• Choosing a project by its stars instead of its last release and its recent commits.
• Counting the license as the whole cost, and leaving out the infrastructure and the hours to run it.
• Collecting rightsizing recommendations nobody applies.
• Changing resources in the cluster or the console that Helm, Argo CD, or Terraform manage, so the next sync puts the old values back.
• Running one tool per cloud with no shared schema, when FOCUS lets the bills meet in one table.

When does a managed platform fit better?

With the tools above, you choose where the software runs and who keeps it current, and several sell a hosted version. A managed platform fits when you want cloud, SaaS, and AI spend in one place, and the savings shipped for you.

LevelFour, which we make, reads your Cloud, SaaS, and AI bills in one view, finds the savings, and ships each one: applied in your cloud or SaaS account with access scoped to that one change, as a pull request to your infrastructure code, or as manual steps. Savings Audit confirms each saving on the bill. LevelFour isn't open source, but its CLI is. Cloud Savings costs 30% of realized savings, and plans start at $30 a month.

See what LevelFour can save you in 5 minutes, with read-only access.

Frequently asked questions